A dataset for evaluating multiple ellipse fitting methods of densely connected contours

Version 2:
The Mathematica source codes "MultiEllipseFittingCodes.zip" are added in this version.

Version 1:
The dataset is made up of an experimental image and a set of synthetic images.
The experimental image is an infrared handprint image (Hand1Raw.png).
The synthetic images include one contour image containing 12 ellipses (12Raw.bmp), one contour image containing 20 ellipses (20Raw.bmp), eight contour images containing 4 ellipses each (4ARaw.bmp, 4BRaw.bmp, ... ,4HRaw.bmp), and a projected Ford circle image (FordDiskFitted.png).
The file "Hand2MedianFiltered.png" is the median-filtered version of "Hand1Raw.png".
And it can be further transformed to a contour image file "Hand3Thresholded.png" by thresholding.
The x-y coordinates of the contours can be saved as csv files.
The files "4A.csv" - "4H.csv" correspond to "4ARaw.bmp, 4BRaw.bmp, ... ,4HRaw.bmp".
"12.csv" and "20.csv" correspond to "12Raw.bmp" and "20Raw.bmp" respectively.
"FordDisk.csv" corresponds to "FordDiskFitted.png".
"Hand01.csv" - "Hand07.csv" correspond to the seven parts of contour in file "Hand3Thresholded.png".
Our goal is evaluating the multiply ellipse fitting framework.
Multiple ellipses extracted from these contours can be saved as files like "...Fitted.png".
Specifically, "4AFitted.png" - "4HFitted.png" correspond to "4ARaw.bmp, 4BRaw.bmp, ... ,4HRaw.bmp".
"12Fitted.png" and "20Fitted.png" correspond to "12Raw.bmp" and "20Raw.bmp" respectively.
"FordDiskFitted.png" corresponds to "FordDiskFitted.png".
"HandFitted.png" corresponds to "Hand3Thresholded.png".
Finally, all above-mentioned files have been compressed into three zip files separately.
The "RawPictures.zip" contains all raw image files.
The "ContourData.zip" contains all csv files.
The "RawPictures.zip" contains all fitted ellipse files.

本版本中新增了 Mathematica 源代码集 "MultiEllipseFittingCodes.zip"。

在版本 1 中，数据集由一张实验图像和一系列合成图像构成。
实验图像为一幅红外手印图像（Hand1Raw.png）。
合成图像包括包含 12 个椭圆的轮廓图像（12Raw.bmp）、包含 20 个椭圆的轮廓图像（20Raw.bmp）、包含每个图像 4 个椭圆的八张轮廓图像（4ARaw.bmp, 4BRaw.bmp, ..., 4HRaw.bmp）以及一张投影福特圆图像（FordDiskFitted.png）。
文件 "Hand2MedianFiltered.png" 是 "Hand1Raw.png" 的中值滤波版本。
该文件可通过阈值化进一步转换为轮廓图像文件 "Hand3Thresholded.png"。
轮廓的 x-y 坐标可以保存为 csv 文件。
文件 "4A.csv" - "4H.csv" 分别对应 "4ARaw.bmp, 4BRaw.bmp, ..., 4HRaw.bmp"。
"12.csv" 和 "20.csv" 分别对应 "12Raw.bmp" 和 "20Raw.bmp"。
"FordDisk.csv" 对应 "FordDiskFitted.png"。
"Hand01.csv" - "Hand07.csv" 分别对应文件 "Hand3Thresholded.png" 中的七个轮廓部分。
本研究的目的是评估多重椭圆拟合框架。
从这些轮廓中提取的多重椭圆可以保存为诸如 "...Fitted.png" 的文件。
具体而言，"4AFitted.png" - "4HFitted.png" 分别对应 "4ARaw.bmp, 4BRaw.bmp, ..., 4HRaw.bmp"。
"12Fitted.png" 和 "20Fitted.png" 分别对应 "12Raw.bmp" 和 "20Raw.bmp"。
"FordDiskFitted.png" 对应 "FordDiskFitted.png"。
"HandFitted.png" 对应 "Hand3Thresholded.png"。
最终，上述所有文件均分别压缩成三个 zip 文件。
"RawPictures.zip" 包含所有原始图像文件。
"ContourData.zip" 包含所有 csv 文件。
"FittedEllipses.zip" 包含所有拟合椭圆文件。